Professional Documents
Culture Documents
Measures Induced by Units PDF
Measures Induced by Units PDF
Abstract. The half-open real unit interval (0, 1] is closed under the ordinary
multiplication and its residuum. The corresponding infinite-valued proposi-
tional logic has as its equivalent algebraic semantics the equational class of can-
cellative hoops. Fixing a strong unit in a cancellative hoop —equivalently, in
the enveloping lattice-ordered abelian group— amounts to fixing a gauge scale
for falsity. In this paper we show that any strong unit in a finitely presented
cancellative hoop H induces naturally (i.e., in a representation-independent
way) an automorphism-invariant positive normalized linear functional on H.
Since H is representable as a uniformly dense set of continuous functions on its
maximal spectrum, such functionals —in this context usually called states—
amount to automorphism-invariant finite Borel measures on the spectrum. Dif-
ferent choices for the unit may be algebraically unrelated (e.g., they may lie
in different orbits under the automorphism group of H), but our second main
result shows that the corresponding measures are always absolutely continuous
w.r.t. each other, and provides an explicit expression for the reciprocal density.
1. Introduction
· 0) satisfying
A cancellative hoop is an algebra (H, +, −,
commutative monoid identities for +, 0,
x−· x = 0,
· y) −
(x − · z=x− · (y + z),
· · y),
x + (y − x) = y + (x −
· y = x.
(x + y) −
The above identities have manifold appearances, and can be seen from various
points of view.
First of all, they characterize the equational theory of truncated addition. In-
deed, the class CH of cancellative hoops coincides with the equational class gen-
· 0), where x −
erated by (R≥0 , +, −, · y = max{0, x − y}. Therefore, they suffice to
capture all identities valid in the positive reals. Also, cancellative hoops are pre-
cisely the positive cones of lattice-ordered abelian groups (`-groups for short), the
latter being the abelian groups (G, +, −, 0) endowed with a lattice order (G, ∧, ∨)
on which the group translations act as order automorphisms. Actually, more is
true: the assignment G 7→ G≥0 = {g ∈ G : g ≥ 0} is functorial, and provides a cat-
egorical equivalence between the class of all `-groups (which is generated by R, and
is equational in the language +, −, 0, ∧, ∨), and CH [6, Theorem 1.17]. Note that
· x) and
the lattice operations are definable in the hoop language by x ∨ y = x + (y −
Key words and phrases. cancellative hoop, lattice-ordered abelian group, MV-algebra, strong
unit, uniform distribution, Cesàro mean.
2010 Math. Subj. Class.: 03G25; 06F20; 11K06.
1
2 G. PANTI, D. RAVOTTI
The many-valued logical system defined by the above identities has the interval
(0, 1] as its set of truth-values, and is called the logic of cancellative hoops [11],
[26], [21, §6.4]; it is what remains of Hajek’s product logic [15] once the falsum is
removed.
A third facet of the matter arises from the observation that the addition opera-
tion + and its residuum − · capture the process of production and consumption of
resources in a Petri net under the firing of a transition. In the classical setting the
resources at each place of a Petri net are marked by an element of Z≥0 , and this
setting is readily generalizable to arbitrary cancellative hoops which are called, in
this setting, Petri algebras. See [1] and references therein.
An element of the cancellative hoop H whose multiples eventually dominate any
element of H is traditionally called a strong unit or, sometimes, an order unit.
Since we are not dealing with the related notion of weak unit, we will simply say
unit. So, explicitly, a unit is an element u ∈ H such that for every h ∈ H there
exists a positive integer k with h ≤ ku. An element u of a hoop of the form C(X)≥0
is a unit iff u(x) 6= 0 for every x ∈ X (remember that X is compact Hausdorff by
definition). Units arise in all contexts cited above: they are archimedean elements
in the `-group enveloping H, gauge scales for falsity in falsum-free product logic,
and universal bounds in bounded Petri nets. Upon defining x ⊕ y = u ∧ (x + y),
¬x = u − x, x y = ¬(¬x ⊕ ¬y), the interval Γ(H, u) := {h ∈ H : h ≤ u} is an MV-
algebra, namely the equivalent algebraic semantics of Lukasiewicz infinite-valued
logic. In the following we assume some familiarity with `-groups, cancellative hoops
and MV-algebras, referring to [5], [6], [8], [11], [24] for all unproved claims.
Let u be a unit in the cancellative hoop H. A state on the pair (H, u) is a
monoid homomorphism m : H → R≥0 such that m(u) = 1; states on (H, u) corre-
spond in a 1–1 canonical way to states on the unital `-group enveloping H and to
states on the MV-algebra Γ(H, u) [22, Theorem 2.4]. We are naturally interested
in automorphism-invariant states, namely those states m such that m ◦ σ = m for
every automorphism σ of H that fixes u. This is a quite natural requirement, since
MEASURES INDUCED BY UNITS 3
F (e) = w = (α1 · · · αn+1 )tr ∈ W1 . Then e(hi ) = αi , and e[H] is the intersection
of R≥0 with the subgroup of R generated by α1 , . . . , αn+1 . Therefore e is discrete
of denominator b iff the subgroup of R generated by α1 , . . . , αn+1 is b−1 · Z iff
bα1 , . . . , bαn+1 ∈ Z and the group they generate is all of Z iff bα1 , . . . , bαn+1 are
relatively prime integers iff w is a rational point of denominator b. This establishes
the bijection between (a) and (b). To every rational point w ∈ W1 there corresponds
the unique primitive point den(w)w ∈ Cone(W ), and u den(w)w = den(w) ·
u(w) = den(w), so (b) and (c) are in bijection.
We need some further tools from piecewise-linear topology: see [29], [34], [12]
for full details. A rational polytope S is the convex hull of finitely many points
of Qn+1 ; its (affine) dimension is the maximum integer d = dim(S) such that S
contains d + 1 affinely independent points. The affine subspace A = aff(S) of Rn+1
spanned by S is then d-dimensional and defined over Q (i.e., it is the 0-set of finitely
many affine polynomials with rational coefficients). A face of S is S itself or the
intersection of S with an hyperplane π such that S is entirely contained in one of
the two closed halfspaces determined by π; the empty set is a face, and every face
different from S is proper. The index of S is the least integer b ≥ 1 such that
bA ∩ Zn+1 6= ∅ [3, p. 518], and the relative interior of S, relint(S), is the topological
interior of S in A or, equivalently, the set of all points of S which do not lie in a
proper face.
Definition 2.4. Let S, A, b be as above. If d = 0, then S = A = {w} for some w ∈
Qn+1 , and we define νA (S) = 1. Assume d > 0. Then there exists a (nonunique)
affine isomorphism ψ : bA → Rd that maps bA ∩ Zn+1 bijectively to Zd . Let us
abuse language by writing b : A → bA for the map w 7→ bw. We define the relative
volume form ΩA on A as
1
ΩA = d+1 · the pull-back via ψ ◦ b of the standard
b
volume form dx̄ = dx1 ∧ · · · ∧ dxd on Rd .
This amounts to saying that, for every continuous function f : A → R with compact
support (more generally, every Riemann-integrable function), we have
Z Z
1
f ΩA = d+1 f ◦ b−1 ◦ ψ −1 dx̄. (1)
A b R d
S3
S1
S4
S2
S5
by mapping w to 0 and v = (2, 1, 1) to 1. The line segment [w, v] has thus νaff(S2 ) -
√ √
measure 1 and λ1 -measure 5, so that νaff(S2 ) = ( 5)−1 λ1 .
The affine span of S3 does not contain points in Z3 , and neither does 2 aff(S3 ). On
the other hand, 3 aff(S3 ) is the line passing through w = (1, 0, 3) and v = (1, 1, 3),
so the index of S3 is 3; we can take ψ : 3 aff(S3 ) → R as the unique affine map that
sends w to 0 and v to 1. By (2) we have
length of ψ[3S3 ]
νaff(S3 ) (S3 ) = .
9
One easily checks that ψ[3S3 ] is the interval [9/5, 3] in R and concludes that
νaff(S3 ) (S3 ) = 2/15; since λ1 (S3 ) = 2/5, we have νaff(S3 ) = 3−1 λ1 .
Finally, νaff(S4 ) = νaff(S5 ) = λ0 by definition.
If d > 0 and b = 1 then νaff(S) (S) is the relative volume of S [2, §5.4]. More
generally we have the following lemma.
Lemma 2.6. Let d > 0 and let S be a d-dimensional simplex, not necessarily
rational, whose vertices w1 , . . . , wd+1 lie on {a = 1}, for some a ∈ Hom(Zn+1 , Z).
Assume that the subspace V spanned by w1 , . . . , wd+1 in Rn+1 is defined over Q,
and let m1 , . . . , md+1 be a Z-basis for the free Z-module Zn+1 ∩ V . Let M ∈
GLd+1 R be defined by (w1 · · · wd+1 ) = (m1 · · · md+1 )M . Then for every affine
function f : S → R we have
|det(M )| f (w1 ) + · · · + f (wd+1 )
Z
f dνaff(S) = .
S (d + 1)!
R
Proof. It is obvious that S f dλ, where λ is any multiple of the d-dimensional
Lebesgue measure, is the product of λ(S) and the average value of f over the
vertices of S. Since νaff(S) is such a multiple, we just need to check the stated
formula for f = 1l, namely
|det(M )|
νaff(S) (S) = . (3)
d!
Writing b for the least positive integer such that b aff(S) contains an integer point,
we have by definition νaff(S) (S) = b−(d+1) νaff(bS) (bS), so everything boils down to
proving
|det(bM )|
νaff(bS) (bS) = . (4)
d!
Now, the strip V ∩ {0 < a < 1} does not contain integer points and hence nei-
ther does the strip V ∩ {0 < a < b}. It follows that Zn+1 ∩ V has a Z-basis
(r1 , . . . , rd+1 ) with r1 , . . . , rd ∈ V ∩ {a = 0} and rd+1 ∈ V ∩ {a = b} = aff(bS). Let
(bw1 · · · bwd+1 ) = (r1 · · · rd+1 )R; then R ∈ GLd+1 R has last row (1 · · · 1).
As in Example 2.5 we define ψ : aff(bS) → Rd by sending rd+1 to 0 and ri + rd+1
to the i-th element ei of the standard basis of Rd . Thus νaff(bS) (bS) is the ordinary
volume of ψ(bS) in Rd , namely |det(P )|/d!, where P is the (d + 1) × (d + 1) matrix
(see, e.g., [32]) whose last row is (1 · · · 1) and whose upper d × (d + 1) minor P 0 is
defined by
(ψbw1 · · · ψbwd+1 ) = (e1 · · · ed )P 0 .
We claim that P = R. Indeed, let U be the (d + 1) × (d + 1) matrix that has 1
along the main diagonal and the last row, and 0 otherwise. Then
(bw1 · · · bwd bwd+1 ) = (r1 + rd+1 · · · rd + rd+1 rd+1 )U −1 R,
MEASURES INDUCED BY UNITS 9
(ii) Let R ∈ (ΠS )max (l) be as in (i). Then R = S ∩P for some P ∈ Πmax (p), with
l ≤ p. By (i), ΣP is a pure p-dimensional complex, and hence S ∩ P is contained
in some p-dimensional element S 0 ∩ P of ΣP . Therefore (S ∩ S 0 ) ∩ P = S ∩ P is
p-dimensional and l ≥ dim(S ∩ S 0 ) ≥Sp; thus S P ∈ Πmax (l).
max
(iii) Let S ∈ Σ (l). Then S = ΠS = (ΠS )max (l), and by the proof of (ii)
max
every R ∈ (ΠS ) (l) is contained in some P ∈ Πmax (l). This shows the left-to-right
inclusion, and the other inclusion is analogous.
(iv) is immediate from (iii).
Definition 2.10. The dimension of the polytopal set W is the dimension of any
polytopal complex Σ of which W is the underlying set; this makes sense because
of Lemma 2.9(iv). Let 0 ≤ l ≤ d = dim(W ). The l-dimensional support of W
is the set of all hyperplanes of the form aff(S), for some S ∈ Σmax (l). The set
A1 , A2 , . . . , Arl of l-dimensional supporting hyperplanes is finite —possibly empty—
and, by Lemma 2.9(ii), depends on W only. For every l such that W has at least
one l-dimensional supporting hyperplane, and for every Borel subset B of W , define
l
(B) = νA1 A1 ∩ B ∩ Σmax (l) + · · · + νArl Arl ∩ B ∩ Σmax (l) .
S S
νW
l
Since νAi (Ai ∩ Aj ) = νAj (Ai ∩ Aj ) = 0 for i 6= j, one checks easily that νW is a
Borel finite measure on W .
0
Note that νW is the counting measure on the set of isolated points of W . As we
l
will be concerned with the measures νW , for l < d, only in the last section of this
d
paper, we save notation by writing νW for νW .
here the first identity is a definition while the second one, which involves Riemann’s
zeta function ζ, is well known [25, pp. 193–195].
As usual, the meaning of identities α(k) = β(k) + O γ(k) such as (5) above is
that there exists a constant C > 0satisfying |α(k) − β(k)| < Cγ(k) for every k.
Analogously, α(k) = β(k) + o γ(k) means that limk→∞ α(k) − β(k) /γ(k) = 0.
We will need the following fact about Cesàro convergence.
Lemma 3.1. Let α : N → R be bounded, let 0 = n0 < n1 < n2 < · · · be a strictly
increasing sequence of natural numbers, and let β ∈ R. Then:
MEASURES INDUCED BY UNITS 11
(i) if
1 X
lim α(t) = β,
k→∞ nk − nk−1
nk−1 <t≤nk
then
1 X
lim α(t) = β; (6)
k→∞ nk
t≤nk
Proof. (i) is due to Cauchy [7, p. 378], and (ii) is [19, Lemma 2.4.1].
By the Ehrhart theory [31, §4.6.2], [2, Theorem 3.23], LbS1 (h) is a quasipolynomial
of degree d and finite period e, with e dividing the lcm of the denominators of the
vertices of bS1 . More precisely, LbS1 (h) has the form
LbS1 (h) = νaff(bS1 ) (bS1 )hd + ld−1 (h)hd−1 + · · · + l1 (h)h + l0 (h),
with ld−1 (h), . . . , l0 (h) rational numbers that only depend on the residue class of h
modulo e. Fix a number M such that |lj (h)| ≤ M for every 0 ≤ j < d and 1 ≤ h.
We thus obtain
X t X
PbS1 (t) = µ νaff(bS1 ) (bS1 )hd + lj (h)hj
h j
h|t
XX t
j
= νaff(bS1 ) (bS1 )ϕd (t) + µ lj (h)h
j
h
h|t
= νaff(bS1 ) (bS1 )ϕd (t) + o ϕd (t) ,
because
1 X X t j dM X t d−1
µ lj (h)h ≤ µ h ,
ϕd (t) j h ϕd (t) h
h|t h|t
12 G. PANTI, D. RAVOTTI
So we get
P P
Ξ(S, ck) Φd (ck/b) t≤ck/b o ϕd (t) t≤ck/b o ϕd (t)
− νaff(bS1 ) (bS1 ) = ≤ P ,
Φd (ck) Φd (ck) Φd (ck) t≤ck/b ϕd (t)
and the last expression goes to 0 as k goes to infinity by Lemma 3.1(i). Finally,
due to the asymptotic estimate (5),
Φd (ck/b) 1
lim νaff(bS1 ) (bS1 ) = νaff(bS1 ) (bS1 ) d+1 = νaff(S1 ) (S1 ).
k→∞ Φd (ck) b
Then:
(i) for every f ∈ C(W ), the Cesàro average A(f, ȳ) exists;
(ii) the functional A(−, ȳ) : C(W ) → R is linear, positive (i.e., A(f, ȳ) ≥ 0 if
f ≥ 0), and normalized (i.e., A(1lW , ȳ) = 1). By the Riesz representation
theorem, A(−, ȳ) is induced by integration w.r.t. a uniquely defined Borel
probability measure on W ;
(iii) if, for every T , the average A(1lT , z̄) exists and equals A(1lT , ȳ), then
A(f, z̄) = A(f, ȳ) for every f ∈ C(W ).
Proof. Since by hypothesis the Cesàro averages along ȳ exist for all characteristic
functions 1lT , they exist for all step functions, i.e., all R-linear combinations of such
characteristic functions. The proof is now straightforward from the fact that the
set of step functions is dense in L∞ (W ) (the Banach space of all bounded functions
on W with the topology of uniform convergence).
The following is our main result.
Theorem 4.2. Let H be a finitely presented cancellative hoop in which a unit u
has been fixed. Let e1 , e2 , e3 , . . . be an enumeration without repetitions of all the
discrete extremal states of (H, u) according to nondecreasing denominators. Then
for every h ∈ H the limit
1X
mu (h) = lim et (h)
k→∞ k
t≤k
exists, does not depend on the enumeration, and the function mu : H → R≥0 thus
defined is an automorphism-invariant state.
MEASURES INDUCED BY UNITS 13
Proof. Let n ≥ 0 be the least integer such that H can be generated by n+1 elements
h1 , . . . , hn+1 .
Remark 4.3. We do not require h1 , . . . , hn+1 ≤ u. In particular, the MV-algebra
Γ(H, u) is still finitely generated —a simple application of the Riesz decomposition
property [14, Proposition 2.2]—, but the least cardinality of a generating set may be
larger than n + 1. For example, let H = Free2 (CH), and let u = x1 ∨ (−kx1 + 1) for
some integer k ≥ 1. Then n = 1, because H is not generable by a single element.
On the other hand, Eu contains k + 2 discrete states of denominator 1, namely
the states h 7→ h(w)/u(w) for w ∈ {0, 1, 1/2, 1/3, 1/4, . . . , 1/(k + 1)}. The usual
representation of Γ(H, u) as an MV-algebra, i.e., as a quotient of Freem (MV) for
some integer m, assumes that u is represented by the constant function 1l. Now,
the only points in the unit cube [0, 1]m that correspond to states of denominator 1
w.r.t. the unit 1l are the 2m vertices of the cube. In our case we need k + 2 such
points, hence presenting Γ(H, u) as an MV-algebra requires at least log2 (k + 2)
generators.
Our H can then be presented as hx1 , . . . , xn+1 ; f1 (x̄) = g1 (x̄), . . . , fr (x̄) = gr (x̄)i.
As cited in §2, Freen+1 (CH) ' M ([0, 1]n )≥0 . Also, W = {w ∈ [0, 1]n : fi (w, 1 −
(w1 ∨ · · · ∨ wn )) = gi (w, 1 − (w1 ∨ · · · ∨ wn )) for every i} is a rational polytopal
set of dimension 0 ≤ d ≤ n, and the hoop homomorphism ρ : H → M (W )≥0
defined by ρh1 = x1 W , . . ., ρhn = xn W , ρhn+1 = 1l − (x1 ∨ · · · ∨ xn ) W is
a McNaughton representation. We adopt the notation of Definition 2.2.
The case d = 0 is trivial. Indeed, in this case W is just a finite set of rational
points, and by Lemma 2.1(ii) the given enumeration is finite, say e1 , e2 , . . . , er .
Therefore, mu is always defined and is a state. Given any automorphism σ of H
such that σ(u) = u, the sets {e 1 , e2 , . . . , er } and {e1 ◦ σ, e2 ◦ σ, . . . , er ◦ σ} are equal,
and hence mu (h) = mu σ(h) for every h.
Let now d ≥ 1; by Lemma 2.3, the enumeration e1 , e2 , e3 , . . . corresponds to
an enumeration w1 , w2 , w3 , . . . of all primitive points in Cone(W ) according to
nondecreasing values of u. Let wt0 = wt /u(wt ) be the point of intersection of W1
with the ray R≥0 wt ; by Lemma 2.3 et (h) = h(wt0 ), so we must show that
1X
mu (h) = lim h(wt0 ) (7)
k→∞ k
t≤k
exists.
Since h W1 is a continuous function, we can use Lemma 4.1(i) to prove the
convergence of (7). Let then T be a rational polytope contained in W1 . S Us-
ing [29, 2.8(6)] we can construct a polytopal complex Σ such that W1 = Σ
and T is a union of elements of Σ; we can thus safely assume T ∈ Σ. Note
that we do not need that Σ is a simplicial complex, nor that it satisfies any
unimodularity condition (see [23] for unimodular simplicial complexes). Let c
be the lcm of the d-indices of the d-dimensional polytopes in Σ. The set of
primitive points in Cone(W ) is then partitioned in blocks B1 , B2 , B3 , . . ., where
Bk = {w ∈ Cone(W ) : w is primitive and c(k − 1) < u(w) ≤ ck}. For every k ≥ 1,
let nk be such that wnk ∈ Bk and wnk +1 ∈ Bk+1 . Generalizing the notation in
Lemma 3.2, let Ξ(W, t) be the number of primitive points in Cone(W ) ∩ {u ≤ t}.
Then X
Ξ(W, ck) = {(S) · Ξ(S, ck) : S ∈ Σ},
14 G. PANTI, D. RAVOTTI
note that νW1 (T ) = 0 if dim(T ) < d. From (8) it is immediate that limk→∞ (nk −
nk−1 )/nk = 0, so we apply Lemma 3.1(ii) and conclude
1X νW1 (T )
lim 1lT (wt0 ) = . (10)
k→∞ k νW1 (W1 )
t≤k
Corollary 4.4. Adopt the hypotheses of Theorem 4.2 and the relative notation.
Then, for every h ∈ H, we have
Z
1
mu (h) = h W1 dνW1 .
νW1 (W1 ) W1
Proof. If d = 0, then νW1 is the counting measure on the finite set W1 =
{w10 , . . . , wr0 }, which is in bijection with {e1 , . . . , er }; since et (h) = h(wt0 ), our
statement is immediate. If d ≥ 1 then, by Lemma 4.1(ii) applied to W1 , mu (h) is
given by integration of h W1 w.r.t. a certain Borel finite measure on W1 . By (10),
that measure is νW1 /νW1 (W1 ).
6. Absolute continuity
Let H be a separating subhoop of some C(X)≥0 (i.e., the inclusion map of H
in C(X)≥0 is a representation). Assume that H, as an abstract hoop, is finitely
presentable. Let u1 , u2 ∈ H be two arbitrary units; we are not assuming any
structure either on X (besides being a compact Hausdorff space), or on the elements
of H (besides being continuous functions). Note that u1 and u2 may lie in different
orbits under the action of the automorphism group of H; equivalently, the MV-
algebras Γ(H, u1 ) and Γ(H, u2 ) may not be isomorphic. This is the case, e.g., of
the functions 1l and u of Remark 4.3, which are both units in M ([0, 1])≥0 .
Let m1 , m2 be the automorphism-invariant states determined by Theorem 4.2.
By Lemma 2.1(iv) there exist two uniquely determined Borel finite measures µ1 , µ1
on X such that Z
mi (h) = h dµi ,
X
16 G. PANTI, D. RAVOTTI
We devote the rest of this section to the proof of Theorem 6.1. The case d = 0 is
X = {p1 , · · · , pr }, in bijection with
trivial. Indeed, in this case X is a finite set, say
Eui , for i = 1, 2, via pj 7→ h 7→ h(pj )/ui (pj ) . All extremal states are discrete, so
mi (h) = r−1 1≤j≤r h(pj )/ui (pj ). The measure µi is therefore
P
1 X 1
µi = δp ,
r ui (pj ) j
1≤j≤r
where δpj is the Dirac unit measure at pj , and the formula (11) is clear.
Assume now d ≥ 1; we shall prove that there exists a constant C such that
Z Z d+2
u1
f dµ2 = C f· dµ1 , (12)
X X u2
for every continuous function f : X → R. As in the proof of Theorem 4.2,
we construct a McNaughton representation ρ : H → M (W )≥0 for a certain d-
dimensional polytopal set W ⊆ [0, 1]n . We adopt the notation in the proof of The-
orem 4.2; in particular, for i = 1, 2, ui is the homogeneous correspondent of ρui , and
Wi = Cone(W ) ∩ {ui = 1}. We thus have two representations ρi : H → C(Wi )≥0 ,
given by ρi h = h Wi . We construct a rational polytopal complex Σ having W as
its underlying set and such that both ρu1 and ρu2 are affine on each polytope of Σ.
ThenSΣi = {Si : Si = Cone(S) ∩ Wi , for S ∈ Σ} is a rational polytopal complex
and Σi = Wi . Let S ∈ Σ be d-dimensional, let bi be the index of Si , and choose
affine isomorphisms ψi as in Definition 2.4 from aff(bi Si ) to Rd . We display our
data in a diagram
b2 S o b2
SO 2 / W2
O 2 O
ψ2 R2
G P P
}
Rd o b1 S 1 o S1 / W1 /X
ψ1 b1 R1
Lemma 6.2. Let p be a point in the interior of ψ1 b1 S1 ; then the absolute value
jF (p) of the determinant of the Jacobian matrix of F at p has value
d+1
b2
jF (p) = .
u2 (ψ1−1 p)
Proof. Let T ⊆ ψ1 b1 S1 be a d-dimensional simplex, not necessarily rational,
containing p in its interior. By [30, Theorem 7.24], jF (p) is the limit, for
T shrinking to p, of the ratio vol(F T )/ vol(T ). By definition, that ratio is
νaff(b2 S2 ) (Gψ1−1 T )/νaff(b1 S1 ) (ψ −1 T ). Let V be the (d + 1)-dimensional real vec-
tor space spanned by S inside Rn+1 . Since V is defined over Q, it intersects Zn+1
in a free Z-module of rank d + 1; let m1 , . . . , md+1 be a Z-basis for this module.
Let w1 , . . . , wd+1 be the vertices of ψ1−1 T and let (w1 · · · wd+1 ) = (m1 · · · md+1 )M
for a certain M ∈ GLd+1 R. Then νaff(b1 S1 ) (ψ −1 T ) = |det M |/d! by Lemma 2.6.
Since the vertices of Gψ1−1 T are b2 w1 /u2 (w1 ), . . . , b2 wd+1 /u2 (wd+1 ), and the
argument above applies to νaff(b2 S2 ) as well, one immediately computes
Letting T shrink to p, the w’s converge to ψ1−1 p and our claim follows.
So (12) is proved with C = D1 /D2 . The expression for C in (ii) follows immediately
by taking f = u2 in (12). The proof of Theorem 6.1 is thus complete.
as above such that relint(S) is not contained in the closure of B. It is then enough
to show the following:
(A) Let U be an open set in W 0 such that w ∈ U ⊆ B, and write M for the
boundary of U in W 0 . Then M ∩ relint(S) contains a point at which the
local dimension of M is not ≤ l − 2.
Let U be as in (A) and let Ū be its closure in W 0 . Then relint(S) is the disjoint
union of relint(S) ∩ U , relint(S) ∩ M , and relint(S) \ Ū . By construction there exist
points p ∈ relint(S) ∩ U and q ∈ relint(S) \ Ū . It follows that every continuum
(i.e., closed connected set) C ⊆ relint(S) containing p and q must intersect M ,
for otherwise (C ∩ U ) ∪ (C \ Ū ) would be a nontrivial disconnection of C. By
Mazurkiewicz’s Theorem [10, Theorem 1.8.19] applied to the open region relint(S)
inside aff(S) ' Rl , the topological dimension of M ∩ relint(S) is not ≤ l − 2.
Therefore there exists a point as required by (A).
l l
S νW of Definitionl 2.10. Note that, since νW is sup-
We shallSneed the measures
ported on Σmax (l), and Σmax (l) \ W l is a νW -nullset, we have
Z Z Z
l l l
f dνW = S f dνW = f dνW ,
W Σmax (l) Wl
for every Riemann-integrable function f : W → R.
Assume now the hypotheses and the notation of Theorem 4.2; we thus have a
McNaughton representation ρ : H → M (W )≥0 and a homeomorphism F : Eu →
W1 = Cone(W ) ∩ {u = 1} as in Lemma 2.3. By Lemma 7.1, for every 0 ≤ l ≤ d =
dim(Eu ), F restricts to a bijection between Eul = {e ∈ Eu : the local dimension of
Eu at e is l} and W1l . Let 0 ≤ l1 < l2 < · · · < lq = d be the dimensions at which
Euli 6= ∅. For each 1 ≤ i ≤ q, let ei1 , ei2 , ei3 , . . . be an enumeration without repetitions
of all the discrete states in Euli according to nondecreasing denominators. Choose
a weight vector p = (p1 , . . . , pq ) (i.e., p ∈ Rq≥0 and
P
pi = 1).
Theorem 7.2. For every i = 1, . . . , q and every h ∈ H the limit
1X i
mlui (h) = lim et (h) (14)
k→∞ k
t≤k
exists and does not depend on the enumeration. The function mpu : H → R≥0
defined by X
mpu (h) = pi mlui (h) (15)
i
is an automorphism-invariant state, which is faithful (i.e., h 6= 0 implies mpu (h) 6= 0)
iff pi > 0 for every i.
Proof. Every convex combination of automorphism-invariant states is clearly an
automorphism-invariant state, so it suffices to show that every mlui is such a state.
Fix then 0 ≤ l ≤ d such that Eul 6= ∅, Sand say that W is the underlying set
of the polytopal complex Σ. Let W = Σmax (l); then W is an l-dimensional
rational polytopal set which, by Lemma 2.9(iii), depends on W and l only. The
restriction map ρh 7→ ρh W is a homomorphism from M (W )≥0 to M (W )≥0 , and
the latter hoop is finitely presented because W is rational. Every element of the
enumeration el1 , el2 , el3 , . . . provided by the hypotheses corresponds to a point of W ,
and can thus be seen as a discrete extremal state of M (W )≥0 , with preservation
of denominators. Moreover, all discrete extremal states of M (W )≥0 appear in the
MEASURES INDUCED BY UNITS 21
q
!
X pi X
li
= li
νW (S)h(S) , (18)
i=1
νW 1
(W1 ) S∈∆max (l ) 1
i
where h(S) is the arithmetical average of h over the vertices of S. All terms pi ,
li li
νW 1
(W1 ), and νW 1
(S) in (18) are > 0. Since h 6= 0 implies h > 0 on at least one
vertex of at least one S ∈ ∆, we have mpu (h) > 0, as desired.
For the reverse implication, let pi = 0 for some 1 ≤ i ≤ q, and let ∆ be a
rational simplicial complex supported on W1 . Since Euli 6= ∅, ∆max (li ) contains a
simplex, say S. Using elementary algebra we construct a Z-basis m1 , . . . , mli +1 of
Zn+1 ∩ (the R-subspace spanned by S) such that the ray R≥0 mj intersects S in its
relative interior, for every j. Let T be the intersection of S with the cone spanned
positively by m1 , . . . , mli +1 ; let also w = m1 + · · · + mli +1 . Since S ∈ ∆max (li ) and
T is contained in the relative interior of S, the simplex T has empty intersection
with every element of ∆ different from S. It readily follows (compare with Mundici’s
theory of Schauder hats [24, §5.3]) that the function h : Cone(W1 ) → R≥0 defined
by:
• h = 0 outside of Cone(T );
22 G. PANTI, D. RAVOTTI
• h(m1 ) = · · · = h(mli +1 ) = 0;
• h(w) = 1;
• for each 1 ≤ i ≤ li + 1, h is linear on the cone spanned positively by
m1 , . . . , mi−1 , w, mi+1 , . . . , mli +1 ;
is continuous and piecewise-linear with integer coefficients. Let h be the unique
element of H such that h is the homogeneous correspondent of ρh. Then h 6= 0
and, since pi = 0 and h = 0 outside Cone(T ), we have mpu (h) = 0 by (17), so mpu
is not faithful.
Remark 7.3. In the specific case u = 1l and p = (1/q, . . . , 1/q), the state mp1l
of Theorem 7.2 coincides with the state s of [23, Theorem 4.1], provided that
MaxSpec H does not contain isolated points. Indeed, let h ∈ H = M (W )≥0
for some d-dimensional rational polytopal set W ⊆ [0, 1]n ; since u = 1l we
have W1 = W and h W1 = h. Choose a unimodular complex ∆ supported
on W (see [23] for all relevant definitions), such that h is affine on S each sim-
plex of ∆. Let T be an l-dimensional unimodular simplex contained in ∆max (l)
(not necessarily belonging to ∆), and let w1 , . . . , wl+1 be the vertices of T . By
the unimodularity assumption (den(w1 )w1 · · · den(wl+1 )wl+1 ) is a Z-basis for
Zn+1 ∩ (the R-subspace spanned by T ). Therefore by Lemma 2.6
l 1
νW (T ) = νaff(T ) (T ) = .
l! den(w1 ) · · · den(wl+1 )
l
Since the unimodular simplexes generate the Borel σ-algebra, νW is the measure λl
of [23, Theorem 2.1]. Therefore the identity (18), which in our specific case reads
q
!
p 1X 1 X
li
m1l (h) = li
νW (S)h(S) ,
q i=1 νW (W ) S∈∆max (l)
agrees with the expression (4) on [23, p. 542] for the state s.
If W contains isolated points, say W 0 = {w1 , . . . , wr } =
6 ∅, then the situation is
slightly different. Indeed, the measure µ0 corresponding to our m01l is
1 X
µ0 = δwj ,
r
1≤j≤r
0
while the measure τ determined by (4) on [23, p. 542] is
X −1 X
1 1
τ0 = · δw .
den(wj ) den(wj ) j
1≤j≤r 1≤j≤r
In general µ0 and τ 0 are different, but they both are automorphism-invariant, since
every automorphism of (H, 1l) must act on W 0 in a denominator-preserving way.
For example, let again W = S1 ∪· · ·∪S5 be the polytopal set of Example 2.5, S4 =
{v}, S5 = {w}. Then den(v) = 2, den(w) = 7, W 0 = {v, w}; take h = x1 W to
be the projection on the first coordinate. Then
Z
1
h dµ0 = h(v) + h(w)
W 2
!
1 1 2 11
= + = ,
2 2 7 28
MEASURES INDUCED BY UNITS 23
while
Z !−1 !
0 1 1 h(v) h(w)
h dτ = + +
W 2 7 2 7
!
14 1 2 19
= + = .
9 4 49 42
The hidden reason for the rigidity of the invariant measure on the higher-
dimensional parts of the spectrum, together with its relative flexibility on the
0-dimensional part, is subtle: the 0-dimensional part is made of finitely many iso-
lated discrete extremal states, while the higher-dimensional parts contain infinitely
many discrete extremal states. Now, in a finite average the input data can be rear-
ranged arbitrarily without affecting the final result, a fact which is definitely false
for infinite averages. Actually, a classical result by Descovich (but dating back to
von Neumann for the real unit interval [19, Theorems II.4.4 and III.2.5]) guarantees
that every regular Borel measure on a compact Hausdorff space with no isolated
points can be realized as in Lemma 4.1(ii) by enumerating appropriately the ele-
ments of any given countable everywhere dense subset. In our case, the set of all
discrete extremal states is such a subset, and in Theorem 4.2 the key constraint on
the resulting measure is the denominator-nondecreasing condition, which becomes
irrelevant precisely on the 0-dimensional part of the spectrum. As that condition
depends crucially on the choice of a unit, it is apparent that units and measures
are inextricably intertwined.
References
[1] E. Badouel, J. Chenou, and G. Guillou. An axiomatization of the token game based on Petri
algebras. Fund. Inform., 77(3):187–215, 2007.
[2] M. Beck and S. Robins. Computing the continuous discretely. Undergraduate Texts in Math-
ematics. Springer, 2007.
[3] M. Beck, S. V. Sam, and K. M. Woods. Maximal periods of (Ehrhart) quasi-polynomials. J.
Combin. Theory Ser. A, 115(3):517–525, 2008.
[4] W. M. Beynon. Duality theorems for finitely generated vector lattices. Proc. London Math.
Soc., 31(3):114–128, 1975.
[5] A. Bigard, K. Keimel, and S. Wolfenstein. Groupes et anneaux réticulés, volume 608 of
Lecture Notes in Math. Springer, 1977.
[6] W. J. Blok and I. M. A. Ferreirim. On the structure of hoops. Algebra Universalis, 43(2-
3):233–257, 2000.
[7] T. J. Bromwich. An introduction to the theory of infinite series. Macmillan and Co., 1908.
Available at the Open Library, http://openlibrary.org/books/OL7073755M.
[8] R. Cignoli, I. D’Ottaviano, and D. Mundici. Algebraic foundations of many-valued reasoning,
volume 7 of Trends in logic. Kluwer, 2000.
[9] A. Dvurečenskij. Subdirectly irreducible state-morphism BL-algebras. Arch. Math. Logic,
50(1-2):145–160, 2011.
[10] R. Engelking. Dimension theory. North-Holland, 1978.
[11] F. Esteva, L. Godo, P. Hájek, and F. Montagna. Hoops and fuzzy logic. J. Logic Comput.,
13(4):531–555, 2003.
[12] G. Ewald. Combinatorial convexity and algebraic geometry. Springer, 1996.
[13] M. Fedel, K. Keimel, F. Montagna, and W. Roth. Imprecise probabilities, bets and functional
analytic methods in Lukasiewicz logic. To appear in Forum Mathematicum, 2009.
[14] K. R. Goodearl. Partially ordered abelian groups with interpolation. American Mathematical
Society, Providence, R.I., 1986.
[15] P. Hájek. Metamathematics of fuzzy logic, volume 4 of Trends in logic. Kluwer, 1998.
24 G. PANTI, D. RAVOTTI
Department of Mathematics, University of Udine, via delle Scienze 206, 33100 Udine,
Italy